Using disaggregated grid cells for a study on the onset of African Civil Wars

Jan Ketil Rød, NTNU, Trondheim
Halvard Buhaug, NTNU, Trondheim

Empirical studies on the onset or duration of civil armed conflicts tend to be studied using country level measures, like for instance GDP per capita, primary commodities as percentage of GDP, type of regime, percentage of mountainous terrain within the country. Civil armed conflicts, however, do seldom extend the total area of a country but tend rather to take place in small regions where government control is hindered and/or where rebel groups have access to contrabands. These are factors that vary geographically. Since both the phenomena being studied and the factors that may explain them are sub-national, any statistical study of civil armed conflicts that applies country-level approximations is potentially flawed. In this paper, we present a disaggregated study that may be an alternative or supplement to country-level investigations. We use a “gridded” approach where grid cells are the unit of observation. For each grid cell, GIS is used to identify whether or not the cell represent a regions of peace or conflict and to generate sub-national measures on key explanatory variables. We use grid of three various resolutions (100 x 100, 150 x 150 and 200 x 200 km) in order to control the robustness of regression coefficient on various aggregations. The units show very strong cross-sectional correlation – that is, the likelihood of conflict for any grid is largely conditional on the conflict involvement of contiguous units. To account for spatial autocorrelation, we include a spatially lagged dependent variable based on first, second and third order neighbourhood.

[Full paper in PDF]